Manus Resumes Independent Operations: What Changed After the Meta Split

Manus formally resumed independent operations on September 1, 2026 — completing the unwind of Meta''s December 29, 2025 acquisition. The founding team continues, public messaging says some users needed backup and restore during the transition, and the stated roadmap shifts toward agents embedded in workflows and proactive agents rather than one-off task runs.

If you lived through August''s backup deadlines, this is the post-transition chapter. If you did not, start with explainx.ai''s August guide: Manus splits from Meta — what gets deleted and how to back up.

TL;DR — what people are asking

Timeline — from acquisition to independence

The awkward middle — "please back up before we delete" while corporate structure unwound — is the part users felt directly. Independence day is cleaner narratively than operationally: teams that restored successfully should spot-check task ordering, artifacts, and billing, not assume byte-identical parity.

Why the Meta deal collapsed — regulatory context

The September 1 announcement closes a thread that started months earlier. On December 29, 2025, Meta closed its acquisition of Manus as part of a broader push into agentic AI. Reuters and Caixin reported the deal at roughly $2 billion, though neither company published final terms. Manus had already moved operations from China to Singapore, downsized its domestic team, and ceased services inside China before the acquisition — a footprint that made cross-border scrutiny predictable.

On April 27, 2026, China's National Development and Reform Commission announced that the foreign-investment security review mechanism had prohibited the transaction and ordered the parties to revoke the acquisition. That is the legal forcing function behind August's data deletion: regulators treating post-acquisition user data as tainted for certain jurisdictions, not a product decision Manus could negotiate away.

explainx.ai flagged the geopolitical angle in July's China overseas AI restrictions coverage — before users received the August backup email. Independence removes Meta as corporate parent; it does not reset the underlying lesson that agent platforms sit at the intersection of product policy and national security review.

Who was affected — and what still applies post-Sep 1

Manus's help center was explicit about scope: "affected regions only" saw subscription sales paused, in-app deletion notices, and the two-day outage. Unaffected users got a confirmation notice — but silence was ambiguous if you signed up via Apple or Facebook without an email on file. Post-independence, Manus states that users who backed up but have not yet restored can still use the restoration portal with no published recovery deadline — a meaningful change from the August urgency window.

Per Manus documentation, deletion covered account information, subscription records, task data, generated artifacts, and connector authorizations for accounts in scope between December 29, 2025 and 7:59 a.m. SGT on August 23, 2026. Data predating the acquisition date was outside the boundary. That acquisition-date cut is the tell: this was legal unwinding of a transaction, not a generic storage migration like Gemini CLI's Antigravity transition.

Storage geography also mattered for risk assessment: Manus stated data resides in the US and Singapore, with details in its Trust Center — relevant when comparing sandbox and egress models in Google Cloud agent sandbox guidance.

How Manus fits the general-agent landscape now

Manus rose in 2025 as a general-purpose agent — browser use, multi-step tasks, limited human steering — briefly tagged in industry press as a "second DeepSeek" moment for agents rather than models. Independence returns Manus to that product lane without Meta's distribution, but also without Meta's balance sheet.

The comparison that matters for builders is not "which agent is smartest" but where state lives. Coding harnesses anchor on git; Manus anchors on task graphs inside Manus. Until export schemas mature, heavy users should treat deliverables as files in storage they control, not as artifacts that only exist inside the agent UI — the same thesis as what is an agent harness.

Leaked system prompts (including Manus) in explainx.ai's AI tools system prompt leak roundup also showed how general agents encode tool policies and refusal boundaries — useful context when Manus shifts toward proactive initiation, where prompt-level guardrails meet real-world blast radius.

What "embedded workflows" and "proactive agents" likely mean

Manus has not shipped a full public spec sheet with this announcement, but the phrases fit industry direction:

Embedded workflows — agents that live inside CRMs, docs, ticketing, and browser tabs instead of a standalone portal. That is the same pressure driving MCP connectors, Claude Cowork-style desktop embedding, and Copilot Studio IT agents.

Proactive agents — systems that initiate work from schedules, inbox signals, or monitoring — not only respond to typed prompts. The harness question becomes permissions and blast radius, not chat UX — see agent harness guide and destructive command guards.

Manus made TIME''s 2026 AI list via founder Xiao Hong — coverage here — so the next phase will be watched as a general-agent product test, not a niche experiment.

Proactive agents — what changes technically

Reactive agents wait for a prompt; proactive agents subscribe to signals — calendar events, inbox rules, cron schedules, monitoring webhooks — and initiate work. That shift moves the security model from "what can this session do?" to "what can this identity do while I'm not watching?"

Builders evaluating Manus's roadmap should ask:

1. Initiation scope — Can the agent start tasks only from user-defined triggers, or from model-inferred "helpful" nudges?
2. Spend caps — Credits, API calls, and third-party purchases need hard ceilings; proactive loops without caps recreate destructive command guard problems in non-coding domains.
3. Human-in-the-loop defaults — Which action classes require approval before execution vs after-the-fact audit?
4. Cross-app OAuth — Embedded workflows mean long-lived tokens in CRMs, email, and ticketing; August's connector-authorization deletion showed how fragile those links are during corporate transitions.

Embedded workflows also push Manus toward the same connector economy as MCP — but MCP standardizes tool reach, not task memory. A Manus agent embedded in Notion still needs a story for where plan state persists when the user switches vendors.

Portability — the lesson that outlives the Meta chapter

August''s deletion boundary taught a durable rule: agent task history is an asset class without a standard export format.

Compare mature portability:

MCP standardized how agents reach out to tools; it did not standardize the record left behind. Until interchange exists, your leverage equals the quality of the vendor''s export button — the thesis of the August Manus backup post.

Geopolitical acquisitions amplify the risk — explainx.ai flagged Meta/Manus investigations in July before August''s backup deadline. Independence reduces one class of policy shock; it does not eliminate product sunset, pricing, or feature deprecation shocks (Gemini CLI → Antigravity migration is the same user experience, different cause).

What to verify if you restored data

1. Task chronology — restores that flatten timelines break audit use cases.
2. Artifacts outside Manus — client deliverables should live in storage you control, not only inside task exports.
3. Billing and plan tier — "welcome back" bonuses were promised in August; reconcile statements.
4. OAuth sign-in paths — Apple/Facebook sign-up users often had no email on file; confirm account linkage still works.
5. Integration permissions — reconnect MCP/tool auth if tokens expired during downtime.

Restore fidelity — common failure modes

Users who completed backup and restore should still expect semantic drift, not just missing rows:

If you missed the August backup window, Manus support is the only path — treat in-platform history from the acquisition window as gone unless support confirms otherwise.

Agent-platform migration playbook (for any vendor)

Manus is a case study, not an exception. Any team betting on a hosted agent should run this checklist before the crisis email:

1. Define the deletion boundary — What corporate event (M&A, region block, sunset) would wipe your history?
2. Export cadence — Weekly export of task summaries and artifacts to git or object storage you control.
3. Identity hygiene — Maintain an email on file; OAuth-only accounts failed August notifications silently.
4. Deliverable rule — Client-facing outputs live outside the agent UI on day one.
5. Runbook for 48-hour outages — Manus proved agents can go dark entirely; have a manual fallback documented.

This is the operational mirror of loop engineering for coding agents: loops fail when state is opaque; general agents fail when history is not portable.

What builders should watch next

Limitations — what independence does not fix

Independent operations solve corporate structure, not every user concern:

• Pricing and credits — August promised "welcome back" bonuses for affected users; long-term pricing post-Meta is unsettled.
• Feature parity — Anything built on Meta-internal integration during the acquisition window may not survive the unwind.
• China market — Manus had already exited China before Meta; independence does not reopen that market.
• Regulatory recurrence — Future cross-border deals or data-localization rules could trigger another boundary event; export habits remain the hedge.

Founders Xiao Hong, Tao Zhang, and Yichao "Peak" Ji continue leading the company per Manus's September announcement — continuity of vision, not guarantee of roadmap delivery dates.

What people are still asking (September 2026)

Can I still restore if I backed up in August but never clicked restore?
Manus's September messaging says yes — restoration portal remains available with no stated deadline. Do not treat that as permanent; export to your own storage anyway.

Is Manus still owned by Meta?
No. The acquisition unwind is complete as of September 1, 2026; NDRC's April prohibition forced the reversal.

Will my data be deleted again?
No second deletion wave was announced for independence day itself. Future policy changes would require new notice — but August proved notices can arrive with weeks, not quarters, of lead time.

Should I trust Manus for production workflows?
That depends on your portability posture. If task history is disposable and deliverables live elsewhere, yes with eyes open. If the agent is your system of record, August was the counterexample.

Related on explainx.ai

• Manus Meta separation — backup and deletion guide (August 2026)
• China overseas AI restrictions — Meta/Manus context
• What is an agent harness?
• What is MCP?
• AI tools system prompts leaked — including Manus
• TIME100 AI 2026 — Manus founder listed
• Google Cloud agent sandbox egress lessons
• Destructive command guard for coding agents
• Loop engineering for coding agents

Official sources: Manus A Note to Our Users (August 2026 transition); Service Change Overview (backup/deletion scope); September 1, 2026 independent-operations announcement via company channels and Caixin Global reporting on the Meta deal collapse.

Independence status, roadmap phrases, and transition details reflect Manus public communications as of September 1, 2026. Backup eligibility, restore fidelity, and feature availability may differ by account — verify in-app and with Manus support for authoritative status.